Deep Learning for Improving Numerical Weather Prediction of Heavy Rainfall. (23rd March 2022)

• Record Type: Journal Article
• Authors: Hess, Philipp; Boers, Niklas
• Is Part Of: Journal of advances in modeling earth systems. Volume 14:Number 3(2022)
• Page Start: n/a
• Page End: n/a
• Publication Date: 2022-03-23
• Subjects: deep learning -- numerical weather prediction -- rainfall extremes; Geological modeling -- Periodicals; Climatology -- Periodicals; Geochemical modeling -- Periodicals; 551.5011
• DOI: 10.1029/2021MS002765
• Languages: English
• ISSNs: 1942-2466

Abstract

Abstract: The accurate prediction of rainfall, and in particular of the heaviest rainfall events, remains challenging for numerical weather prediction (NWP) models. This may be due to subgrid‐scale parameterizations of processes that play a crucial role in the multi‐scale dynamics generating rainfall, as well as the strongly intermittent nature and the highly skewed, non‐Gaussian distribution of rainfall. Here we show that a U‐Net‐based deep neural network can learn heavy rainfall events from a NWP ensemble. A frequency‐based weighting of the loss function is proposed to enable the learning of heavy rainfall events in the distributions' tails. We apply our framework in a post‐processing step to correct for errors in the model‐predicted rainfall. Our method yields a much more accurate representation of relative rainfall frequencies and improves the forecast skill of heavy rainfall events by factors ranging from two to above six, depending on the event magnitude. Plain Language Summary: Modeling rainfall is challenging because of its large variability in space and time, and its highly skewed distribution. Numerical weather prediction (NWP) models have to be simulated on discretized grids with finite resolution. Although important especially for the generation of rainfall, small‐scale processes can therefore not be resolved explicitly and must be paremeterized, that is, included as empirical functions of the resolved variables. This introduces model biases that can lead to an under‐ or overestimation of heavy rainfall events. Here we apply a deep neural network (DNN) to correct biases in the rainfall forecast of a NWP ensemble. The DNN is optimized with a loss function that includes weights to account for heavy rainfall events, and shows substantially improved performance in their prediction. Key Points: Correcting biases in the rainfall forecast of a numerical weather prediction ensemble with a deep neural network Training with a weighted loss function combining two terms enables the neural network to learn the heavy tailed target distribution The method improves the relative frequency and categorical skill scores of heavy rainfall